Control device of internal combustion engine

Patent metadata

Field	Value
Publication number	US-11047325-B2
Application number	US-201916552162-A
Country	US
Kind code	B2
Filing date	Aug 27, 2019
Priority date	Sep 14, 2018
Publication date	Jun 29, 2021
Grant date	Jun 29, 2021

Abstract

A control device of an internal combustion engine includes: a parameter value acquiring part acquiring values of input parameters; a computing part utilizing a model using a neural network to calculate a value of an output parameter, and a control part controlling operation of the engine. The model outputs the value of the output parameter from an output layer node if the values of the input parameters are input to the input layer nodes. When an abnormality occurs at values of part of the input parameters among the input parameters, the computing part uses the corrected model to calculate the value of the output parameter, the corrected model being provided by correcting the model so that a value changing in accordance with a value of an abnormal input parameter is not input from a input layer node corresponding to the abnormal input parameter to a hidden layer node.

First claim

The invention claimed is:

1. A control device of an internal combustion engine, the control device including a digital computer and a memory, the digital computer being configured to:
acquire values of input parameters showing an operating state of the internal combustion engine;
utilize a model using a neural network which comprises a plurality of layers including an input layer, a hidden layer, and an output layer to calculate a value of an output parameter when the values of the input parameters are input to the input layer; and
control operation of the internal combustion engine based on the value of the output parameter calculated by utilizing the model, wherein
the memory stores a plurality of relationships of the values of all of the input parameters and the value of the output parameter as sets of training data,
the model is configured to output the value of the output parameter from a node of the output layer when the values of the input parameters are input to nodes of the input layer, each of the nodes of the input layer receiving the value of a different one of the input parameters, and
the digital computer is configured so that occurrence of an abnormality in one of the input parameters causes the digital computer to (1) obtain a corrected model which is a version of the model in which a constant output is provided (i) from one of the input nodes that receives the one of the input parameters in which the abnormality occurs (ii) to nodes of the hidden layer, (2) use the sets of training data stored in the memory to train the corrected model, thereby obtaining a trained corrected model, and (3) utilize the trained corrected model to calculate the value of the output parameter.

2. The control device according to claim 1 , wherein the constant output from the one of the input nodes that receives the one of the input parameters in which the abnormality occurs is zero.

Assignees

• Toyota Motor Co Ltd

Inventors

• Maeda Shingo

Classifications

• G06N3/048

  Activation functions · CPC title

• F02D41/2487

  Methods for rewriting · CPC title

• F02D41/2438

  Active learning methods · CPC title

• G05B13/027

  using neural networks only · CPC title

• F02D41/222

  relating to the failure of sensors or parameter detection devices · CPC title